Device abnormality detection system and method

Abstract

The present invention discloses an equipment anomaly detection system. The system obtains data samples that differ from the mathematical expectation of samples in a data set of a collected device, sets them as abnormal points and no-difference points, excludes them from the sample set to obtain image features, uses directional gradient histogram, local binary pattern and gray-level co-occurrence matrix to extract the image features, intelligently classifies graphic samples formed by motor types, gradually loads each motor from light load to rated load or overload to obtain three-phase current signals, establishes a fault model according to the circuit structure in the equipment, reconstructs the fault signal through an observer to achieve fault reconstruction, establishes a switching system mathematical model of the converter and a partial fault model of parametric faults, compares the residual evaluation function of the observed value and the actual value with a threshold, and detects a fault if the threshold is exceeded to improve the accuracy and timeliness of equipment anomaly detection and ensure the safety of equipment operation.

Description

Technical Field

[0001] The present invention relates to the field of equipment detection technology, and more particularly to an equipment anomaly detection system and method. Background Art

[0002] With the rapid development of science and technology, the power electronics systems of equipment are becoming increasingly complex. The reliable operation of power electronics systems is crucial in engineering practice. A system failure can disrupt stable operation, affecting various functions and causing economic losses. Inverters, sensors, and other components are currently part of most equipment. Component failures, such as aging, are common, and are often concentrated in semiconductor devices or sensor failures within the circuit, making detection difficult and inaccurate. Summary of the Invention

[0003] In view of this, the present invention provides a device abnormality detection system and method, which establishes a mathematical model of the switching system of the converter and a partial fault model of parametric faults, compares the residual evaluation function of the observed value and the actual value of the trip with a threshold, and detects a fault if the threshold is exceeded to improve the accuracy and timeliness of device abnormality detection. The following technical solutions are specifically adopted to achieve this.

[0004] In a first aspect, the present invention provides a device anomaly detection system, comprising:

[0005] A preprocessing module is used to set data samples in the data set of the collected device that are different from the sample mathematical expectation as abnormal points and no difference points, and exclude them from the sample set to obtain image features;

[0006] Image feature extraction module, used to extract image features using oriented gradient histogram, local binary pattern and gray level co-occurrence matrix;

[0007] An intelligent classification module is used to intelligently classify the graphical samples formed by motor types, gradually loading each motor from light load to rated load or overload to obtain three-phase current signals;

[0008] The anomaly detection module is used to establish a fault model based on the circuit structure within the device, convert sensor faults into actuator faults through a first-order filter, introduce performance indicators to obtain sufficient conditions for the existence of the matrix gain of the sliding mode observer, solve the linear matrix inequality to obtain the gain matrix of the observer, compare the residual signal generated by the observer with the set threshold, and detect the fault when the residual is greater than the threshold. When the fault is detected, the adaptive sliding mode observer is automatically activated, and the fault signal is reconstructed through the observer to achieve fault reconstruction.

[0009] As a further improvement to the above technical solution, the adaptive synovial observer is automatically activated when a fault is detected, including:

[0010] The threshold is calculated by the residual function in the sensor failure within the device, where the residual is generated between the measured value and the observed value. The residual of the inductor current and capacitor voltage are expressed as: Determine the total residual function, then When the system is detected to be fault-free, the residual signal r(t) is a very small value; when a sensor fault is detected, the residual r(t) will gradually increase;

[0011] The threshold is selected according to the residual evaluation function, and the expression of the residual evaluation function is: Where 0 is the initial evaluation time, M is the evaluation time, and the threshold is selected based on the upper limit of the residual signal when the circuit is fault-free. Fault detection is performed by comparing the residual function and the threshold. The detection process is:

[0012] As a further improvement to the above technical solution, when the fault detection observer detects a fault, it automatically activates an adaptive sliding mode observer to reconstruct the fault signal, including:

[0013] Reconstruct the fault signal, and the expression of the adaptive sliding mode observer is x(t)=A σ(t) x(t)+B σ(t) u(t)-G L e y (t)+V,y(t)=Cx(t), where the nonlinear term V is Where ρ0 is a constant, and it is guaranteed that ρ0>ρ, ρ (t) is the adaptive rate of the observer, the adaptive rate ρ (t) The expression for ρ is (t) =θsign(e yi (t)-δ), which is based on e yi (t) changes with the difference between it and the custom parameter δ, the custom parameter δ is a smaller value, and the defined parameter θ is a larger value;

[0014] Assume that e(t)=x'(t)-x(t), e y (t) = y'(t) - y(t) to obtain the error state system, e' = (A σ(t) -G L C)e(t)-Df(t)+V, when the observer state converges to the system state, the sensor fault of the system is tracked.

[0015] As a further improvement of the above technical solution, the image feature extraction module includes an acquisition unit and a control unit. The acquisition unit is used to sample and transform the amplified noise signal and calculate the circuit noise power spectrum density. The noise signal amplified by the low-noise amplifier is a connected analog signal. The ADC driven by the control unit performs signal sampling and analog-to-digital conversion. All data are read from the data table, the independent variables and dependent variables are extracted, and the parameters are reorganized and standardized to meet the standards for model training. The data is used as the feature value of deep learning and the data set is segmented into a training set and a test set. The test set is used to input the training function.

[0016] As a further improvement of the above technical solution, the preprocessing module performs a data preprocessing process based on the same state estimation data obeying a non-standard normal distribution, and eliminates samples that fall outside two standard deviations as outliers and no-difference points, and according to the number of eliminated samples, the original sample set is expected to be filled into the corresponding sample set to complete the samples.

[0017] As a further improvement of the above technical solution, the data is cleaned, segmented, normalized and standardized, outliers in the sample set are excluded and the mean is filled. The original sampling data includes training samples and test samples. The training samples are input into the host computer software to train the classification model, and the test data is input into the existing diagnostic model. The diagnostic model is used to predict the category of the new sample, the accuracy of the test sample is statistically predicted and data analysis is performed.

[0018] As a further improvement of the above technical solution, the Hall sensor in the device determines whether a fault occurs by its own signal state. When the Hall sensor does not fail, the angle information output by the Hall position sensor is used as the expected position, and the error between the position signal output by the resolver and the expected position signal is determined to determine whether the resolver fails. The expression of the judgment condition is |p res -p h |>△p, where p res is the position signal output by the resolver, p h is the position signal of the Hall position sensor, and △p is the fault detection threshold.

[0019] As a further improvement of the above technical solution, after the resolver fails, the position sensor is switched to the Hall sensor, and the discrete position signal output by the Hall position sensor is used to perform square wave control on the motor in the device. When the Hall position sensor is within a certain Hall interval, the current electrical angle position signal is expressed as Where θ(t) is the current electrical angle position of the motor rotor, θ k is the starting value of the electrical angle of the current Hall interval, t,t nThe total running time of the motor and the time point when it runs to the starting boundary of the current Hall interval, is the motor electrical angular velocity.

[0020] As a further improvement to the above technical solution, the detection is performed using the logic high and low level state information output by the Hall sensor and the edge state information of the Hall, including:

[0021] The preset fault detection variable ES(rj,i) (j=a,b,c; i=a,bc; i≠j) represents the logic level state of Hall sensor i at the rising edge of Hall sensor j, and ES(fj,i) (j=a,b,c; i=a,b,c; i≠j) represents the logic level state of Hall sensor i at the falling edge of Hall sensor j, where a, b, and c represent the output signals of the three Hall sensors A, B, and C, respectively;

[0022] When the Hall sensor is not faulty, ES(fj,i)≠ES(rj,i) in a Hall signal. Under normal circumstances, ES(fb,a)≠ES(rb,a), ES(fc,a)≠ES(rc,a), and the expression is K=|ES(fj,i)-ES(rj,i)|=1. Under normal circumstances, the value of K is always 1.

[0023] In a second aspect, the present invention further provides a method for detecting an abnormality in a device, comprising the following steps:

[0024] The data samples that are different from the mathematical expectation of the samples in the data set of the acquired device are set as abnormal points and no difference points, and are excluded from the sample set to obtain image features;

[0025] Image features are extracted using histogram of oriented gradients, local binary pattern and gray-level co-occurrence matrix;

[0026] Intelligently classify the graphic samples formed by motor types, and gradually load each motor from light load to rated load or overload to obtain three-phase current signals;

[0027] A fault model is established based on the circuit structure within the device. The sensor fault is converted into an actuator fault through a first-order filter. Performance indicators are introduced to obtain the sufficient conditions for the existence of the matrix gain of the sliding mode observer. The gain matrix of the observer is obtained by solving the linear matrix inequality. The residual signal generated by the observer is compared with the set threshold. When the residual is greater than the threshold, the fault is detected. When the fault is detected, the adaptive sliding mode observer is automatically activated, and the fault signal is reconstructed through the observer to achieve fault reconstruction.

[0028] The present invention provides a device anomaly detection system and method. Data samples that differ from the mathematical expectation of samples in a data set of a collected device are set as abnormal points and no-difference points, and are excluded from the sample set to obtain image features. The image features are extracted using a directional gradient histogram, a local binary pattern, and a gray-level co-occurrence matrix. Graphic samples formed by motor types are intelligently classified. Each motor is gradually loaded from a light load to a rated load or overload to obtain a three-phase current signal. A fault model is established according to the circuit structure in the device. A sensor fault is converted into an actuator fault through a first-order filter. A performance index is introduced to obtain a sufficient condition for the existence of a matrix gain of a sliding mode observer. A linear matrix inequality is solved to obtain a gain matrix of the observer. A residual signal generated by the observer is compared with a set threshold. A fault is detected when the residual is greater than the threshold. When a fault is detected, an adaptive sliding mode observer is automatically activated. The fault signal is reconstructed by the observer to achieve fault reconstruction. By establishing the mathematical model of the switching system of the converter and the partial fault model of parametric faults, the residual evaluation function of the observed value and the actual value of the trip is compared with the threshold. If the threshold is exceeded, the fault is detected to improve the accuracy and timeliness of equipment abnormality detection and ensure the safety of equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 This is a structural block diagram of the device anomaly detection system provided by the present invention;

[0031] Figure 2 This is a flow chart of the device anomaly detection method provided by the present invention. DETAILED DESCRIPTION

[0032] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0033] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. Conversely, when an element is referred to as being "directly on" another element, there is no intermediate element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only.

[0034] See Figure 1 The present invention provides a device anomaly detection system, comprising:

[0035] A preprocessing module is used to set data samples in the data set of the collected device that are different from the sample mathematical expectation as abnormal points and no difference points, and exclude them from the sample set to obtain image features;

[0036] Image feature extraction module, used to extract image features using oriented gradient histogram, local binary pattern and gray level co-occurrence matrix;

[0037] An intelligent classification module is used to intelligently classify the graphical samples formed by motor types, gradually loading each motor from light load to rated load or overload to obtain three-phase current signals;

[0038] The anomaly detection module is used to establish a fault model based on the circuit structure within the device, convert sensor faults into actuator faults through a first-order filter, introduce performance indicators to obtain sufficient conditions for the existence of the matrix gain of the sliding mode observer, solve the linear matrix inequality to obtain the gain matrix of the observer, compare the residual signal generated by the observer with the set threshold, and detect the fault when the residual is greater than the threshold. When the fault is detected, the adaptive sliding mode observer is automatically activated, and the fault signal is reconstructed through the observer to achieve fault reconstruction.

[0039] In this embodiment, when a fault is detected, the adaptive sliding film observer is automatically activated, and the threshold is calculated using a residual function. In the case of a sensor fault in the device, the residual is generated between the measured value and the observed value. The residuals of the inductor current and capacitor voltage are expressed as: Determine the total residual function, then When the system is detected to be fault-free, the residual signal r(t) is a very small value; when a sensor fault is detected, the residual r(t) will gradually increase; the threshold is selected according to the residual evaluation function, and the expression of the residual evaluation function is Where 0 is the initial evaluation time, M is the evaluation time, and the threshold is selected based on the upper limit of the residual signal when the circuit is fault-free. Fault detection is performed by comparing the residual function and the threshold. The detection process is:

[0040] It should be noted that when the fault detection observer detects a fault, it will automatically activate an adaptive sliding mode observer to reconstruct the fault signal. The expression of the adaptive sliding mode observer is x(t) = A σ(t) x(t)+B σ(t) u(t)-G L e y (t)+V,y(t)=Cx(t), where the nonlinear term V is Where ρ0 is a constant, and it is guaranteed that ρ0>ρ, ρ (t) is the adaptive rate of the observer, the adaptive rate ρ (t) The expression for (t) =θsign(e yi (t)-δ), which is based on e yi (t) changes with the difference between the custom parameter δ, the custom parameter δ is a smaller value, and the defined parameter θ is a larger value; the preset e(t) = x'(t) - x(t), e y (t) = y'(t) - y(t) to obtain the error state system, e' = (A σ(t) -G L C)e(t)-Df(t)+V, when the observer state converges to the system state, the sensor fault of the system is tracked.

[0041] It should be understood that the image feature extraction module includes an acquisition unit and a control unit. The acquisition unit is used to sample and transform the amplified noise signal and calculate the circuit noise power spectral density. The noise signal amplified by the low-noise amplifier is connected to the analog signal. The control unit drives the ADC to sample the signal and perform analog-to-digital conversion. All data is read from the data table, the independent and dependent variables are extracted, and the parameters are reorganized and standardized to meet the model training standards. This data is used as feature values ​​for deep learning, and the dataset is divided into a training set and a test set. The test set is used as input to the training function. The preprocessing module performs data preprocessing based on the same state estimation data following a nonstandard normal distribution. Samples falling outside two standard deviations are removed as outliers and in-difference points. Based on the number of removed samples, the original sample set is expected to be filled into the corresponding sample set to complete the sample. The observer ensures that the residual meets the robustness index. The linear matrix inequality is solved to obtain the observer gain. The observed value is compared with the actual value, and the residual signal and the expected value are compared. Through the fault detection logic, fault detection is achieved for component parameter faults.

[0042] Optionally, the data is cleaned, segmented, normalized and standardized, outliers in the sample set are excluded and the mean is filled. The original sampling data includes training samples and test samples. The training samples are input into the host computer software to train the classification model, and the test data is input into the existing diagnostic model. The diagnostic model is used to predict the category of the new sample, the accuracy of the test sample is statistically predicted and data analysis is performed.

[0043] In this embodiment, the Hall sensor in the device determines whether a fault occurs by its own signal state. When the Hall sensor does not fail, the angle information output by the Hall position sensor is used as the expected position, and the error between the position signal output by the resolver and the expected position signal is determined to determine whether the resolver fails. The expression of the judgment condition is |p res -p h |>△p, where p res is the position signal output by the resolver, p h is the position signal of the Hall position sensor, and △p is the fault detection threshold.

[0044] It should be noted that after the resolver fails, the position sensor is switched to the Hall sensor, and the discrete position signal output by the Hall position sensor is used to perform square wave control on the motor in the device. When the Hall position sensor is within a certain Hall interval, the current electrical angle position signal is expressed as Where θ(t) is the current electrical angle position of the motor rotor, θ k is the starting value of the electrical angle of the current Hall interval, t,t n The total running time of the motor and the time point when it runs to the starting boundary of the current Hall interval, is the motor electrical angular velocity.

[0045] Optionally, the detection is performed using the logic high and low level state information output by the Hall sensor and the edge state information of the Hall, including:

[0046] The preset fault detection variable ES(rj,i) (j=a,b,c; i=a,bc; i≠j) represents the logic level state of Hall sensor i at the rising edge of Hall sensor j, and ES(fj,i) (j=a,b,c; i=a,b,c; i≠j) represents the logic level state of Hall sensor i at the falling edge of Hall sensor j, where a, b, and c represent the output signals of the three Hall sensors A, B, and C, respectively;

[0047] When the Hall sensor is not faulty, ES(fj,i)≠ES(rj,i) in a Hall signal. Under normal circumstances, ES(fb,a)≠ES(rb,a), ES(fc,a)≠ES(rc,a), and the expression is K=|ES(fj,i)-ES(rj,i)|=1. Under normal circumstances, the value of K is always 1.

[0048] In this embodiment, the fault detection observer meets performance requirements under arbitrary switching signals and zero initial conditions. This performance requirement enhances system robustness and minimizes the impact of faults on the residual signal. The sliding mode observer's observed value and the actual circuit value generate a residual signal. The difference between the residual signal and a threshold and the fault detection logic determine whether the circuit has a fault. When the residual signal exceeds the threshold, a fault is detected, and the adaptive sliding mode observer is used to reconstruct the fault signal.

[0049] See Figure 2 The present invention also provides a device abnormality detection method, comprising the following steps:

[0050] S1: Obtain data from the data set of the collected device that are different from the sample mathematical expectation. The data samples are set as abnormal points and no difference points, and are excluded from the sample set to obtain image features.

[0051] S2: Image features are extracted using histogram of oriented gradients, local binary patterns, and gray-level co-occurrence matrix;

[0052] S3: Intelligently classify the graphic samples formed by motor types, and gradually load each motor from light load to rated load or overload to obtain three-phase current signals;

[0053] S4: A fault model is established based on the circuit structure within the device. The sensor fault is converted into an actuator fault through a first-order filter. Performance indicators are introduced to obtain sufficient conditions for the existence of the matrix gain of the sliding mode observer. The linear matrix inequality is solved to obtain the gain matrix of the observer. The residual signal generated by the observer is compared with the set threshold. When the residual is greater than the threshold, the fault is detected. When the fault is detected, the adaptive sliding mode observer is automatically activated. The fault signal is reconstructed through the observer to achieve fault reconstruction.

[0054] In this embodiment, the state type of the sample is clearly known at the time of data collection. The data samples collected under the same circuit state follow a normal distribution. Before deep learning diagnosis, the data needs to be sorted to improve the quality of the sample set and reduce the adverse effects on the model. During the data cleaning and data preparation process, due to the abnormalities of missing data, sampling errors, and noise during data collection, the data samples are preprocessed according to the model's requirements for data, the types and characteristics of the sample data, and the difficulties in the actual sample collection process. The abnormal points and error points in the samples are screened and filtered to prevent the abnormal points and error points from having adverse effects during the data normalization and standardization process, thereby reducing modeling errors, enabling the model to better identify the intrinsic characteristics of the sample set, and improving the accuracy of model classification or prediction.

[0055] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0056] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not limiting, and thus other examples of the exemplary embodiments may have different values.

[0057] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0058] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that variations and modifications are possible without departing from the scope of the present invention, and such variations and modifications are fully within the scope of protection of the present invention.

Claims

1. A device anomaly detection system, characterized in that: include: A preprocessing module is used to set data samples in the data set of the collected device that are different from the sample mathematical expectation as abnormal points and error points, and exclude them from the sample set to obtain image features; Image feature extraction module, used to extract image features using oriented gradient histogram, local binary pattern and gray level co-occurrence matrix; An intelligent classification module is used to intelligently classify the graphical samples formed by motor types, gradually loading each motor from light load to rated load or overload to obtain three-phase current signals; The anomaly detection module is used to establish a fault model based on the circuit structure within the device, convert sensor faults into actuator faults through a first-order filter, introduce performance indicators to obtain sufficient conditions for the existence of the matrix gain of the sliding mode observer, solve the linear matrix inequality to obtain the gain matrix of the observer, compare the residual signal generated by the observer with the set threshold, and detect the fault when the residual is greater than the threshold. When a fault is detected, the adaptive sliding mode observer is automatically activated, and the fault signal is reconstructed through the adaptive sliding mode observer to achieve fault reconstruction; Automatically activates the adaptive sliding mode observer when a fault is detected, including: The threshold is calculated by the residual function in the sensor failure within the device, where the residual is generated between the measured value and the observed value. The residual of the inductor current and capacitor voltage are expressed as: Determine the total residual function, then When the system is detected to be fault-free, the residual signal r(t) is a very small value; when a sensor fault is detected, the residual r(t) will gradually increase; The threshold is selected according to the residual evaluation function, and the expression of the residual evaluation function is: Where 0 is the initial evaluation time, M is the evaluation time, and the threshold is selected based on the upper limit of the residual signal when the circuit is fault-free. Fault detection is performed by comparing the residual function and the threshold. The detection process is: When the sliding mode observer detects a fault, it automatically activates an adaptive sliding mode observer to reconstruct the fault signal, including: Reconstruct the fault signal, and the expression of the adaptive sliding mode observer is x(t)=A σ(t) x(t)+B σ(t) u(t)-G L e y (t)+V,y(t)=Cx(t), where the nonlinear term V is Where ρ0 is a constant, and it is guaranteed that ρ0>ρ, ρ (t) is the adaptive rate of the observer, the adaptive rate ρ (t) The expression for (t) =θsign(e yi (t)-δ), which is based on e yi (t) changes with the difference between it and the custom parameter δ, the custom parameter δ is a smaller value, and the defined parameter θ is a larger value; Assume that e(t)=x'(t)-x(t), e y (t) = y'(t) - y(t) to obtain the error state system, e' = (A σ(t) -G L C)e(t)-Df(t)+V, when the observer state converges to the system state, the sensor fault of the system is tracked; The Hall sensor in the device determines whether a fault occurs by its own signal status. When the Hall sensor does not fail, the angle information output by the Hall position sensor is used as the expected position. The error between the position signal output by the resolver and the expected position signal is used to determine whether the resolver fails. The expression of the judgment condition is |p res -p h |>Δp, where p res is the position signal output by the resolver, p h is the position signal of the Hall position sensor, Δp is the fault detection threshold; After the resolver fails, the position sensor is switched to the Hall sensor, and the discrete position signal output by the Hall position sensor is used to perform square wave control on the motor in the device. When the Hall position sensor is within a certain Hall interval, the current electrical angle position signal is expressed as Where θ(t) is the current electrical angle position of the motor rotor, θ k is the starting value of the electrical angle of the current Hall interval, t,t n The total running time of the motor and the time point when it runs to the starting boundary of the current Hall interval, is the motor electrical angular velocity; The detection is performed using the logic high and low level status information output by the Hall sensor and the edge status information of the Hall, including: The preset fault detection variable ES(rj,i) (j=a,b,c; i=a,bc; i≠j) represents the logic level state of Hall sensor i at the rising edge of Hall sensor j, and ES(fj,i) (j=a,b,c; i=a,b,c; i≠j) represents the logic level state of Hall sensor i at the falling edge of Hall sensor j, where a, b, and c represent the output signals of the three Hall sensors A, B, and C, respectively; When the Hall sensor is not faulty, ES(fj,i)≠ES(rj,i) in a Hall signal. Under normal circumstances, ES(fb,a)≠ES(rb,a), ES(fc,a)≠ES(rc,a), and the expression is K=|ES(fj,i)-ES(rj,i)|=1. Under normal circumstances, the value of K is always 1.

2. The device anomaly detection system according to claim 1, characterized in that: The image feature extraction module includes an acquisition unit and a control unit. The acquisition unit is used to sample and transform the amplified noise signal and calculate the circuit noise power spectrum density. The noise signal amplified by the low-noise amplifier is a connected analog signal. The ADC driven by the control unit samples the signal and performs analog-to-digital conversion. Among them, all data is read from the data table, the independent variables and dependent variables are extracted, and the parameters are reorganized and standardized to meet the standards for model training. The data is used as the feature value of deep learning and the data set is divided into a training set and a test set. The test set is used to input the training function.

3. The device anomaly detection system according to claim 1, characterized in that: The preprocessing module performs data preprocessing according to the non-standard normal distribution of the same state estimation data, removes samples that fall outside two standard deviations as outliers and no-difference points, and fills the original sample set into the corresponding sample set according to the number of removed samples to complete the samples.

4. The device anomaly detection system according to claim 3, characterized in that: The data is cleaned, segmented, normalized and standardized, outliers in the sample set are excluded and the mean is filled. The original sampling data includes training samples and test samples. The training samples are input into the host computer software to train the classification model, and the test data is input into the existing diagnostic model. The diagnostic model is used to predict the category of new samples, the accuracy of the test samples is statistically predicted and data analysis is performed.

5. A device anomaly detection method according to any one of claims 1 to 4, characterized in that: The following steps are involved: The data samples in the data set of the acquired device that are different from the mathematical expectation of the sample are set as abnormal points and error points, and are excluded from the sample set to obtain image features; Image features are extracted using histogram of oriented gradients, local binary pattern and gray-level co-occurrence matrix; Intelligently classify the graphic samples formed by motor types, and gradually load each motor from light load to rated load or overload to obtain three-phase current signals; A fault model is established based on the circuit structure within the device. The sensor fault is converted into an actuator fault through a first-order filter. Performance indicators are introduced to obtain the sufficient conditions for the existence of the matrix gain of the sliding mode observer. The gain matrix of the observer is obtained by solving the linear matrix inequality. The residual signal generated by the observer is compared with the set threshold. When the residual is greater than the threshold, the fault is detected. When the fault is detected, the adaptive sliding mode observer is automatically activated, and the fault signal is reconstructed through the observer to achieve fault reconstruction.

Images